Aside from these questions of reference panel composition, we can also ask whether the choice of imputation method matters in African datasets. Our HapMap 3 comparisons suggest that Beagle has trouble imputing African genotypes, and we wanted to see if this conclusion would hold up in the MalariaGEN data. To address this, we used Beagle to impute the Gambian validation set using two different reference panels: (i) the GMB panel and (ii) a composite reference set containing the GMB panel, the GHN panel, and the HM3 panels with majority African ancestry (ASW, LWK, MKK, and YRI). We decided to reduce the HM3 set to these panels (which we label “HM3.afr”) in case Beagle’s difficulties in the previous comparisons were caused by the inclusion of non-African haplotypes in the reference panel. For consistency, we also imputed from the GMB+GHN+HM3.afr panel with IMPUTE2.
